 |
|
 |
| |
|
|
 Hong Kong Observatory - SWIRLS Optimizations and Parallelization
The Hong Kong Observatory (HKO) developed an operational nowcasting system, called SWIRLS, to forecast rainfall distribution to support the issuance of warnings for heavy rain in Hong Kong, with new radar data being ingested every 6 minutes. Recently, the execution speed was found to be insufficient for a few very widespread rain situations and such kind of "worst case" scenario was expected to become even more acute when nowcasting needs grow in the near future.
HKO determined that the SWIRLS software needed optimizing and paralleling in order to improve execution performance and allow for continual development, and CLUSTERTECH successfully won the corresponding tender.
In the 3-month project, CLUSTERTECH was tasked to optimize and parallelize two core software modules of SWIRLS with a primary aim to assure that the execution time be made within 6 minutes (an operational time constraint) even in the "worse case" scenario. At the end, this primary aim was fulfilled with a 4.4x speedup over the original program using 6 processors of the SWIRLS operational computer. The parallelized SWIRLS software is currently running in a production environment in HKO.
|
|
|
| |
|
|
HSBC
CLUSTERTECH was engaged by HSBC from 2002 to 2006 to develop software for the analysis and forecasting of capital flows and market directions in foreign exchange. Based on the information captured by the strong global franchise of HSBC, different artificial intelligence modeling techniques were developed and tested to predict financial markets.
We harnessed the power offered by grid computing for CPU and I/O-intensive modeling research and development work. Optimization runs which would have taken a month to run on a single machine were updated within a day via parallelization. CLUSTERTECH also developed a Software Development Kit to modularize different forecasting engines for model validation and evaluation. CLUSTERTECH's software product CPE was the backbone of this parallel framework, which enabled fast turnaround time of model training and optimization.
|
|
|
| |
|
|
Environmental Protection Department (EPD), Hong Kong
CLUSTERTECH was engaged by EPD to redevelop the SAQM simulation software used for pollution modeling. The original SGI Origin code took 48 hours to perform simulation for a 44-hour period and thus could not produce forecasts in real time. In addition, expansion for higher resolution, larger regional coverage, and new chemical processes was not possible.
CLUSTERTECH ported over 40,000 lines of legacy Fortran code and parallelized the application with MPI. Careful verification of the numerical accuracy and performance scaling of the parallel program was also undertaken. The production model can now perform regional simulation in less time, over a wider area and with better resolution and accuracy.
|
|
|
| |
|
|
Pan Asian
CLUSTERTECH was engaged by Pan Asian Mortgage Company Limited in 2004 to develop software for the structuring of mortgage backed securities in the Hong Kong market. Starting from a database of individual mortgage loans, complex cash flow partitioning rules were built into a system that can run scenario analysis and generate various reports on the behavior of each class of mortgage backed security. An object-oriented approach was taken so that pluggable components can be easily replaced to cater for different sets of structuring rules. Advanced features such as parameter backout based upon user specified objectives allow users to improve its capital efficiency.
|
|
|
| |
|
|
Solutions to Clients in Mainland China
We provided High Performance Computing (HPC) cluster solutions that included as a package system design, cluster management software, system implementation, usage training and application benchmark to more than thirty institutions in mainland China. The clients have diverse requirements on number crunching power, interconnect bandwidth and latency, and IO throughput. Our customized and end-to-end solutions ensured the effective utilization of the HPC clusters.
Press Release Samples:
|
|
|
| |
|
|
|
|
